Interesting news for everyone working on paralinguistics and pathological speech. I am very happy to announce that we provide a dataset for a subchallenge of the Computational Paralinguistics ChallengE (ComParE) 2022. The ComParE challenge has long been a success at Interspeech conferences. Due to some changes in the handling of special sessions, the challenge moved on and is now a part of ACM multimedia grand challenge.

We provided KSF-C, which is a derivate of our Kassel State of Fluency dataset (KSoF). KSoF was accepted for presentation and publication at the LREC 2022 conference. You can check out the pre-print from arXiv if you want to know more about the base paper.

KSF-C

The KSoF corpus consists of 3s long speech segments from 37 German speakers with an overall duration of 4.6.h. The segments were annotated by three labellers as belongig to one of 7 classes (block, prolongation, sound repetition, word/phrase repetition, modified speech technique, interjection, no disfluency) and with some additional information, e.g., about the recording quality. Annotators were able to assign more than one label per segment. For the challenge, ambiguously labelled segments wer removed, thus only featuring 4601 segments. The task proposed in this challenge, is the classification of speech segments as one of 8 classes – the seven stuttering-related classes and an eighth garbage class, denoting segments that are unintelligible, contain no speech, or are negatively affected by background noise. The dataset is split into three speaker-independent partitions (Train 23 speakers, Dev 6 speakers, Test 8 speakers.

Additional infos can be obtained from the ComParE challenge website. This includes the baseline paper, including baseline classification results. If you want to participate in the challenge, please contact me at sebastian.bayerl@th-nuernberg.de for the EULA and after signing and submitting the EULA I will give you access to the data on zenodo.

The deadline for submitting a challenge paper is the 18th of June 2022. So it’s plenty of time to find a nice solution for the challenge task.

I would love to see you participate in the challenge and I’m looking forward to hearing from you!